{"id":3744,"date":"2026-07-17T05:30:00","date_gmt":"2026-07-17T05:30:00","guid":{"rendered":"https:\/\/naaia.ai\/?p=3744"},"modified":"2026-07-16T15:19:55","modified_gmt":"2026-07-16T15:19:55","slug":"prohibited-ai-practices-ai-act-compliance","status":"publish","type":"post","link":"https:\/\/naaia.ai\/en\/prohibited-ai-practices-ai-act-compliance\/","title":{"rendered":"Prohibited AI Practices: understanding the restrictions under the AI Act\u00a0"},"content":{"rendered":"\n

The AI Act <\/strong>(Regulation (EU) 2024\/1689) is based on a risk-based classification of AI systems. Within this framework, certain practices are considered unacceptable<\/strong> and are therefore explicitly prohibited. <\/p>\n\n\n\n

These practices are defined in Article 5, which establishes a closed list of prohibited uses, based on their impact on fundamental rights, including decision-making autonomy, equal treatment and privacy. <\/p>\n\n\n\n

Following the adoption of the Regulation, the European Commission proposed adjustments through the AI Omnibus<\/strong>. This proposal introduces new prohibited practices<\/strong>, particularly in relation to certain uses of generative AI, to reflect recent technological developments. <\/p>\n\n\n\n

I \u2013 Subliminal,\u00a0manipulative\u00a0or deceptive techniques\u00a0<\/h2>\n\n\n\n

AI systems that use subliminal techniques<\/strong>, operating below a person\u2019s level of awareness, or deliberately manipulative or deceptive techniques<\/strong>, are prohibited where they have the effect of materially distorting a person\u2019s behaviour<\/strong>. <\/p>\n\n\n\n

The key criterion is the impairment of the person\u2019s ability to make an informed decision, leading them to take a decision they would not otherwise have taken, with a risk of harm. <\/p>\n\n\n\n

Example: an application that exploits cognitive biases to encourage users to make purchases or investments they would not have made under normal circumstances.<\/em> <\/p>\n\n\n\n

II \u2013 Exploitation of vulnerabilities\u00a0<\/h2>\n\n\n\n

AI systems that exploit vulnerabilities linked to age, disability or a specific social or economic situation<\/strong> are prohibited. <\/p>\n\n\n\n

This applies where such vulnerabilities are used to materially distort the behaviour of a person or group and cause, or are likely to cause, harm. <\/p>\n\n\n\n

Example: a system targeting financially vulnerable individuals with personalised offers encouraging them to take out unsuitable credit may fall within this category.\u00a0<\/em><\/p>\n\n\n\n

III \u2013 Evaluation or classification of persons (\u201csocial scoring\u201d)\u00a0<\/h2>\n\n\n\n

AI systems used for the evaluation or classification of natural persons<\/strong> based on social behaviour or personal characteristics are prohibited when they lead to specific adverse outcomes. <\/p>\n\n\n\n

Two situations are covered. First, where such evaluations are used in contexts unrelated to those in which the data was originally generated or collected. Second, where they result in unjustified or disproportionate adverse treatment. <\/p>\n\n\n\n

Example: a score assigned to an individual based on online behaviour and later used to deny access to a service unrelated to that behaviour.<\/em> <\/p>\n\n\n\n

IV \u2013 Risk assessment for criminal offences<\/h2>\n\n\n\n

\u00a0The use of AI systems to assess or predict the risk <\/strong>of a natural person committing a criminal offence<\/strong> is prohibited where this assessment is based solely on profiling or personal characteristics. <\/p>\n\n\n\n

A distinction is made where such systems are used to support a human assessment based on objective and verifiable facts directly linked to criminal activity. <\/p>\n\n\n\n

Example: a tool assigning a criminal risk level based on personal traits or general behavioural data, without any direct link to a specific offence.<\/em> <\/p>\n\n\n\n

V \u2013 Creation of facial recognition databases through untargeted scraping\u00a0<\/h2>\n\n\n\n

AI systems that create or expand facial recognition databases through untargeted scraping of facial images<\/strong> from the internet or CCTV footage are prohibited. <\/p>\n\n\n\n

This prohibition addresses the creation of large-scale databases without control over the origin of the data or the consent of the individuals concerned. <\/p>\n\n\n\n

Example: a system trained on millions of images automatically collected from social media platforms without users\u2019 knowledge.<\/em> <\/p>\n\n\n\n

VI \u2013 Emotion recognition in the workplace and education\u00a0<\/h2>\n\n\n\n

The use of AI systems for emotion recognition<\/strong> is prohibited in specific contexts, namely the workplace and educational institutions<\/strong>. Exceptions apply only where the use is justified for medical or safety purposes. <\/p>\n\n\n\n

Example: a system analysing employees\u2019 facial expressions or voice to assess engagement or stress levels in a workplace setting.<\/em> <\/p>\n\n\n\n

VII \u2013 Biometric categorisation based on sensitive characteristics<\/h2>\n\n\n\n

AI systems for biometric categorisation<\/strong> are prohibited where they are used to infer sensitive attributes<\/strong>, such as race, political opinions, religious beliefs or sexual orientation. <\/p>\n\n\n\n

The objective is to prevent the use of biometric data to analyse or classify protected characteristics. <\/p>\n\n\n\n

Example: a system claiming to infer political opinions from facial features or biometric data.<\/em> <\/p>\n\n\n\n

VIII \u2013 Real-time remote biometric identification in publicly accessible spaces for law enforcement purposes\u00a0<\/h2>\n\n\n\n

The use of real-time remote biometric identification systems<\/strong> in publicly accessible spaces<\/strong> for law enforcement purposes<\/strong> is, generally, prohibited. <\/p>\n\n\n\n

Limited exceptions apply, including the search for missing persons, the prevention of serious and imminent threats, or the identification of suspects in relation to serious criminal offences. These uses are subject to strict requirements, including necessity, proportionality and prior authorisation<\/strong>. <\/p>\n\n\n\n

IX \u2013 New prohibited practices introduced by the AI Omnibus\u00a0<\/h2>\n\n\n\n

The AI Omnibus<\/strong> complements the existing framework by introducing new prohibited uses, particularly in relation to generative AI. These include AI systems capable of generating or manipulating highly sensitive content, such as: <\/p>\n\n\n\n